I work on the structured introduction of AI-based development tools into software teams and on the integration of language models into products. I use these tools every day in my work, Claude Code in particular.
Areas of work
- Adoption of coding agents in teams: defining explicit, verifiable project conventions, automated checks for compliance with them, organizing project documentation, criteria for dividing work between developers and agents, review processes for generated code.
- Training for developers and technical leads, with a dedicated course covering coding agents, automated guardrails, frontier models, and local models.
- Measuring AI use: collection and analysis of telemetry from development sessions, with import of existing history and dashboards per project and per session. It makes it possible to measure AI use, the projects it is concentrated on, and the related costs, without the data leaving the company's infrastructure. The system has been in use in my work since September 2026.
- Integration of models into products: design and development of MCP servers that make an application's data and features accessible to agents, design of APIs intended for use by agents, adoption of locally run models when confidentiality requirements do not allow the use of external providers.
- Privacy: assessment of personal data processing connected to the use of third-party language models, with particular reference to the content of prompts and logs, as part of the privacy and GDPR service.
Method
The method I use, and that I bring to teams, places coding agents within a structured software engineering process, based on:
- written project conventions on architecture, error handling, documentation, and style, which form the context of every agent work session and are updated with the corrections that come up during the project;
- automated checks that verify compliance with the conventions at the end of each agent task and prevent the task from completing when there are violations;
- documentation that describes the current state of the system, alongside a decision log that preserves its history and rationale;
- defining architectural choices before implementation;
- functional verification in the browser at the end of every change to the user interface.
Approach
Coding agents are tools that support the development team. The speed at which they produce code makes the experience needed to decide what to build, and to judge whether the proposed solutions are correct, even more important. My involvement aims to make the team's work more effective; the team remains responsible for the software that goes into production.